Fix nondefault compression handling

[peytondmurray]

This PR fixes compression handling for nondefault compression types, closes #321 and partially addresses #331. Additionally if recreate_dataset is called, we now log a warning message if compression is enabled for the dataset; this is because h5py stores compression settings in Dataset._filters, and we have no way of distinguishing these from other filters that may be present in the file.

In practice I think this is an unusual use case, but we now emit a warning if this does happen.

If an invalid compression setting is requested in write_dataset, a more informative error message is displayed with a listing of all detected compression settings.

[ArvidJB]

Looks good!

Can you add a test for blosc compression via pytables like in #321 (comment) ? pytables should be an optional dependency for testing only.

[peytondmurray]

I parametrized the tests I added to try with both hdf5plugin and pytables, but pytables appears to be incompatible with numpy>=2 (PyTables/PyTables#1172). I've made the tests skip if the user has both pytables and numpy>2 installed. Right now if you want to use blosc compression with the latest numpy, it seems like hdf5plugin is the way to go.